Why can't Google AI Overview spell words correctly?

Have you ever imagined a technology capable of cracking complex codes or solving advanced mathematical problems, yet failing completely when asked to count letters in a word? This phenomenon is now plaguing Google AI Overview. Despite being heralded as the future of search engines, this AI exhibits a fundamental flaw — an inability to spell correctly. According to a recent report dated 28 May 2026, Google AI Overview provided surprising responses to basic queries. For instance, it claimed there are two ‘P’s in ’Google’. It also stated there’s only one ‘r’ in ‘poop’ and misspelled ‘journalism’ as ‘j-o-u-r-n-a-d-i-s-m’. Even for public figures, the AI correctly identified the number of letters but jumbled the order, such as spelling the US President’s surname as ‘t-r-p-u-m’. These errors add to the long list of AI ‘hallucinations’, following previous suggestions for users to put glue on pizza or eat stones based on satirical sources. ‘Counting letters in words is a known challenge for LLMs, and we are working to fix this specific issue,’ a Google representative stated in an official statement cited by TechCrunch on Thursday, 28 May. This issue is not merely a software glitch but a limitation in the fundamental architecture of Large Language Models (LLMs). Here’s the technical reason why AI struggles with spelling: Matthew Guzdial, an AI researcher at the University of Alberta, explained that when AI processes a word, it has a single encoding for its meaning but does not ‘see’ the individual letters. This is why classic challenges like counting ‘r’s in ’strawberry’ often fail even for the most advanced AI models. Despite Google’s ongoing efforts to resolve this, experts like Sheridan Feucht of Northeastern University doubt a ‘perfect tokenisation’ solution exists. This spelling failure serves as a crucial reminder that AI is not omniscient. Human validation remains key when consuming AI-generated information. Always double-check sensitive data provided by AI.